DARPA held its proposers day on April 10 in Arlington, Virginia, for O-CIRCUIT, a programme that asks researchers to build working computers from living neural tissue. The Organoid Cytomorphic Intelligence Resulting from Convergent Understanding and Information Transfer programme, managed by Jeffrey Zaleski at DARPA’s Biological Technologies Office, will fund teams over 42 months to develop biological processing units that can learn, adapt, and compute while consuming a fraction of the energy demanded by silicon chips.
The pitch is straightforward: modern AI systems draw hundreds of watts for tasks that an insect brain handles on microwatts. DARPA’s benchmark is the fruit fly, whose roughly 140,000 neurons consume less than six milliwatt-hours per day while processing visual input, navigating three-dimensional space, and responding to odours. O-CIRCUIT wants engineered biological systems that approach that efficiency while performing useful military tasks.
Two task areas
The programme is split into two tracks. Task Area 1, Architecture, asks teams to build biological processing units with sufficient learning capability to play a video game — DARPA cites Ms. Pac-Man as a reference benchmark — at near-human proficiency while maintaining energy consumption rates comparable to natural neural systems. The work will examine how organoid architecture, cell composition, circuitry, and health affect learning performance, with teams expected to develop maintenance and support systems that keep the living tissue functional over extended periods.
Task Area 2, Action, moves from the screen to the field. Teams will integrate a biological olfactory sensor system with a BPU and mount the combined system on an unmanned platform capable of chemotaxis — navigating toward an odour source. The goal is a drone that can detect tens of different chemical signatures and steer itself toward them using biological sensing rather than conventional electronic sensors. It is the kind of capability that has obvious applications in chemical and explosive detection, though DARPA’s programme description focuses on the engineering challenge rather than specific end uses.
Timeline and structure
O-CIRCUIT is structured as an 18-month first phase followed by two optional 12-month phases, with awards made through Other Transaction agreements. Abstracts are due May 11, and DARPA expects to select teams for oral presentations before awarding first-phase contracts beginning in November 2026.
Context
The programme arrives at a moment when biological computing has moved from science fiction to funded lab work. In March, researchers at Tohoku University published a study in PNAS demonstrating that living rat cortical neurons could be trained using FORCE learning to generate complex target signals in real time — the first closed-loop reservoir computing result using biological tissue. Cortical Labs in Melbourne has been training neural cultures to play Pong since 2022 and raised capital to pursue commercial applications. O-CIRCUIT represents the US defence establishment placing its own bet on the same underlying premise: that engineered biological systems can outperform silicon at specific tasks, particularly where power, weight, and adaptability matter more than raw throughput.
The proposers day drew both academic and defence-industry attendees. Whether any team can produce a biological processor that meaningfully outperforms a conventional edge-computing chip within 42 months remains an open question, but the programme’s existence signals that DARPA considers organoid computing a serious enough prospect to fund at scale.